Differential diagnosis of acid-fast mycobacteria utilizing an artificial intelligence-based approach in histopathological analysis
5 affiliations
- 1Department of Pathology, Hubei Provincial Hospital of Integrated Chinese and Western Medicine, Wuhan, 430000 China
- 2Laboratory Department, Hubei Provincial Hospital of Integrated Chinese and Western Medicine, Wuhan, 430000 China
- 3Data Governance Center, Nanchang People’s Hospital, Nanchang, 330009 China
- 4School of Public Health, Jiangxi Medical College, Nanchang University, Nanchang, 330006 China
- 5Jiangxi Provincial Key Laboratory of Disease Prevention and Public Health, Nanchang University, Nanchang, 330006, China
Abstract
Background and objectives: Tuberculosis is a significant infectious disease and ranks among the top ten causes of mortality globally, with recent estimates indicating that tens of millions have succumbed to it. The primary etiological agent of tuberculosis is the acid-fast positive Mycobacterium tuberculosis (AFB). Traditionally, the diagnosis of AFB in paraffin-embedded human biopsy tissues is achieved through microscopic examination of Ziehl-Neelsen (ZN) stained slides. Due to the necessity of high magnification and the meticulous, repeated searches required by skilled diagnostic physicians, we have developed an artificial intelligence system based on histomorphology to automate the identification of acid-fast positive bacteria.
Methods: This study adopted the VGG-16 architecture from the traditional convolutional neural network and performed fine-tuning on it, forming a set of algorithms required for the classification model. It was combined with histopathological images to construct an intelligent recognition model for anti-acid mycobacteria. The dataset includes cases that are positive and negative for acid-fast staining. The typical inflammatory changes of tuberculosis, such as multinucleated giant cells and necrotic granulomas, can be diagnosed through microscopic observation. Those cases with acid-fast positive mycobacteria are classified as abnormal and assigned a value of 1. For cases without tuberculosis inflammation and no detection of acid-fast positive bacteria, they are classified as normal and assigned a value of 0. The sensitivity, specificity and accuracy of the model are verified through the training set and validation set data. Finally, an automatic identification website window for acid-fast mycobacteria is constructed using the built model. The efficacy of the model is verified through the actual scanning of digital pathological section images.
Results: The model’s prediction accuracy ranges from 0.955 to 0.960, and the area under the curve value was 0.921. The loss function of the model is lower than 0.25, and as the number of iterations increases, the training values get closer and closer to the predicted values. The confusion matrix analysis shows that the model predicts the number of negative acid-resistant color blocks to be 98.32% of the actual negative quantity, and the number of positive acid-resistant color blocks to be 92.50% of the actual value. The 95% confidence interval sensitivity, specificity and accuracy of the model are respectively 0.925, 0.983 and 0.982. It indicates that the constructed model predicts the acid-fast positive mycobacteria values more accurately compared to the actual values.
Conclusion: This study successfully constructed a deep learning classification model for acid-fast mycobacteria using the improved VGG-16 model. The research results were subsequently designed into a web page, which can be used for the auxiliary diagnosis of acid-fast mycobacteria, reducing the workload of pathologists and lowering the error rate.
Related
- Histopathological Imaging and Artificial Intelligence for Cancer Diagnosis and Gleason Grading: A Brief Review of Prostate Cancer.
- Generative Artificial Intelligence Models in Breast Cancer Diagnosis: A Systematic Review
- Artificial Intelligence in Oncologic Imaging: Deep Learning, Radiomics, and Clinical Integration for Precision Cancer Diagnosis
- Artificial Intelligence-Guided Extracellular Vesicle Platforms for Diagnosis and Treatment of Chronic Kidney Disease: A Scoping Review and Future Directions
- Artificial Intelligence-Enhanced Cardiac Magnetic Resonance in Hypertrophic Cardiomyopathy: Diagnosis, Phenotypic Characterization, and Risk Prediction
BibTeX
@article{zhang2026differential,
title = {{Differential diagnosis of acid-fast mycobacteria utilizing an artificial intelligence-based approach in histopathological analysis}},
author = {Shiwei Zhang and Yilin Qu and Fan Wang and Hao Fang and Pan Qin and Libin Deng and Hongliang Ji},
journal = {Diagnostic Pathology},
year = {2026},
month = jul,
doi = {10.1186/s13000-026-01817-4},
url = {https://doi.org/10.1186/s13000-026-01817-4},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Differential diagnosis of acid-fast mycobacteria utilizing an artificial intelligence-based approach in histopathological analysis
AU - Shiwei Zhang
AU - Yilin Qu
AU - Fan Wang
AU - Hao Fang
AU - Pan Qin
AU - Libin Deng
AU - Hongliang Ji
PY - 2026
DA - 2026/07/31
JO - Diagnostic Pathology
DO - 10.1186/s13000-026-01817-4
UR - https://doi.org/10.1186/s13000-026-01817-4
AB - Background and objectives: Tuberculosis is a significant infectious disease and ranks among the top ten causes of mortality globally, with recent estimates indicating that tens of millions have succumbed to it. The primary etiological agent of tuberculosis is the acid-fast positive Mycobacterium tuberculosis (AFB). Traditionally, the diagnosis of AFB in paraffin-embedded human biopsy tissues is achieved through microscopic examination of Ziehl-Neelsen (ZN) stained slides. Due to the necessity of high magnification and the meticulous, repeated searches required by skilled diagnostic physicians, we have developed an artificial intelligence system based on histomorphology to automate the identification of acid-fast positive bacteria. Methods: This study adopted the VGG-16 architecture from the traditional convolutional neural network and performed fine-tuning on it, forming a set of algorithms required for the classification model. It was combined with histopathological images to construct an intelligent recognition model for anti-acid mycobacteria. The dataset includes cases that are positive and negative for acid-fast staining. The typical inflammatory changes of tuberculosis, such as multinucleated giant cells and necrotic granulomas, can be diagnosed through microscopic observation. Those cases with acid-fast positive mycobacteria are classified as abnormal and assigned a value of 1. For cases without tuberculosis inflammation and no detection of acid-fast positive bacteria, they are classified as normal and assigned a value of 0. The sensitivity, specificity and accuracy of the model are verified through the training set and validation set data. Finally, an automatic identification website window for acid-fast mycobacteria is constructed using the built model. The efficacy of the model is verified through the actual scanning of digital pathological section images. Results: The model’s prediction accuracy ranges from 0.955 to 0.960, and the area under the curve value was 0.921. The loss function of the model is lower than 0.25, and as the number of iterations increases, the training values get closer and closer to the predicted values. The confusion matrix analysis shows that the model predicts the number of negative acid-resistant color blocks to be 98.32% of the actual negative quantity, and the number of positive acid-resistant color blocks to be 92.50% of the actual value. The 95% confidence interval sensitivity, specificity and accuracy of the model are respectively 0.925, 0.983 and 0.982. It indicates that the constructed model predicts the acid-fast positive mycobacteria values more accurately compared to the actual values. Conclusion: This study successfully constructed a deep learning classification model for acid-fast mycobacteria using the improved VGG-16 model. The research results were subsequently designed into a web page, which can be used for the auxiliary diagnosis of acid-fast mycobacteria, reducing the workload of pathologists and lowering the error rate.
SN - 1746-1596
ER -